Predicting the causative pathogen among children with osteomyelitis using Bayesian networks - improving antibiotic...

Area of Science:

• * Computational biology and bioinformatics
• * Infectious disease modeling
• * Clinical decision support systems

Background:

• * Osteomyelitis (OM) in children is a serious bacterial infection requiring prompt antibiotic therapy.
• * Current methods for guiding antibiotic selection rely on cultures, which can be slow, inaccurate, or contaminated.
• * Inappropriate antibiotic selection risks treatment failure, toxicity, and antibiotic resistance.

Purpose of the Study:

• * To develop and validate a Bayesian Network (BN) model for predicting causative pathogens in pediatric osteomyelitis.
• * To guide individually targeted antibiotic therapy at the point-of-care.
• * To improve antibiotic selection and patient outcomes in pediatric OM.

Main Methods:

• * Development of a Bayesian Network (BN) model integrating clinical, demographic, and culture data with expert knowledge.
• * Causal inference framework used to model relationships between pathogens, cultures, and patient variables.
• * Model performance evaluated using log loss cross-validation and AUC for predicting culture results.

Main Results:

• * Predicted pathogen prevalence: Staphylococcus aureus (56%), unculturable bacteria (27%), other culturable bacteria (17%).
• * Model achieved AUC values of 0.68-0.77 for predicting culture results.
• * BN-recommended antibiotics were rated optimal or adequate by experts in 82-98% of cases.

Conclusions:

• * Bayesian Network models show robust performance in predicting pathogens and guiding antibiotic selection for pediatric osteomyelitis.
• * This approach facilitates a shift towards individually-targeted antibiotic therapy, improving management of serious bacterial infections.
• * The developed tool has potential for real-time clinical decision support and broader application in pediatric infectious diseases.